Skip to main content
Menu
Close Search

Statistical model uses existing data to categorize patient engagement

perscriptions
Researchers in the Department of Mathematics and Statistics built a statistical model that uses three variables—appointments attended, prescriptions refilled and immunizations recorded—to build a score, which was associated with high, medium or low patient engagement. This model using existing data from electronic health records.

Following doctor’s orders may seem like a no-brainer, but a surprising number of patients fail to do so. About half the time, patients do not take their medications as prescribed, according to the Centers for Disease Control and Prevention.

Health care is transitioning from a fee-for-service to a value-based care industry, according Emily Griese, director of the Sanford Data Collaborative. “We are held responsible, not just for doing the surgery, for example, but for the outcomes that go along with that. Understanding how engaged our patients are with health-care delivery is crucial to predicting outcomes for patients.” 

Assistant Professor Semhar Michael and Associate Professors Xijin Ge and Gemechis Djira of the Department of Mathematics and Statistics worked with former Associate Professor of Pharmacy Practice Surachat Ngorsuraches developed a statistical model that uses existing patient data to evaluate how engaged patients are in their health-care regimen. Public Health and Health Outcomes Data Analyst Patricia Da Rosa and Professor Howard Wey, now retired, from the College of Nursing, were also involved in the 2017 Sanford Data Collaborative research project.

Semhar Michael
Semhar Michael

The Patient Activation Measure is a lengthy patient survey that quantifies patient engagement, but Sanford Health wants to build a similar score using data that is already being collected, Michael explained. In addition, she said, “Because patients assess themselves, there is a lot of subjectivity. We wanted to make it more objective, to look at what they do rather than taking their word for it.”

Michael and her colleagues used a latent variable (engagement) modeling approach through finite mixture models and three variables—appointments attended, prescriptions refilled and immunizations recorded—to build a score, which was associated with high, medium or low patient engagement. They accessed three years of electronic health records data on 147,687 adult patients with two or more chronic conditions.

The scores from the new model were associated with more outcome measurements than those from 1,442 patients given the Patient Activation Measure, Michael reported. “It was a more accurate indicator of patient engagement.”

Michael worked with graduate student Mosa Alsabhi to refine the model. The research was part of Alsabhi’s master’s thesis, which was completed in July 2018. 

“Sanford really liked this scoring and we are investigating adding more variables and trying to use this score to identify patients that need more provider involvement,” Michael explained.

“The score [generated by this predictive algorithm] will calibrate the way we provide services to individuals who have a low engagement potential,” Griese said. Additional support, such as a case manager and online tools, can help improve outcomes for these patients.